Chinese & English Video/Audio to Transcript & SRT

Turn public Chinese (Mandarin) and English videos or podcast audio files into timestamped JSON segments, clean plain-text transcripts, and ready-to-download .srt subtitle files—automatically extracting existing captions when available and falling back to built-in CPU faster-whisper transcription with no external API key required.

---

1. What It Does

Pass a list of public video or audio URLs and get back:

• Timestamped segments ([{start, end, text}]) for search indexing, AI summarization, or subtitle alignment.
• Full plain-text transcript (text) ready for LLM pipelines or content analysis.
• Standard SubRip (.srt) subtitles both inline in the Dataset and saved as individual .srt files in the default Key-Value Store.

It uses a two-stage hybrid pipeline to keep your costs low:

1. Captions Path (source: "captions"): Checks metadata first and extracts existing manual or auto-generated subtitles in seconds without downloading media.
2. Local Whisper Fallback (source: "whisper"): When no captions exist (such as direct podcast .mp3 files or uncaptioned Bilibili videos), downloads only the audio stream and transcribes it on CPU using faster-whisper (compute_type="int8", vad_filter=True). You do not need to bring an OpenAI API key.

---

2. Supported Platforms (Cloud-Verified)

Every platform listed below has been tested and verified on Apify Cloud without cookies or account logins:

Platform	URL Pattern	How It Works on Apify Cloud	Verified Languages
Bilibili (B站公开视频)	https://www.bilibili.com/video/BV...	Resolves metadata and audio via Bilibili public player API and transcribes Mandarin speech via local faster-whisper (source: "whisper"), or extracts public JSON subtitles when exposed without login.	Chinese (Mandarin zh)
Direct Audio & Podcasts	https://.../*.mp3 (e.g. Archive.org, podcast RSS audio CDNs, .m4a, .wav)	Downloads the audio stream directly, enforces maxDurationMinutes truncation, and transcribes speech via faster-whisper (source: "whisper").	Chinese (zh), English (en), and 90+ languages
TED Talks	https://www.ted.com/talks/...	Extracts official human-translated WebVTT subtitles directly (source: "captions") in ~1–2 seconds per talk.	English (en), Chinese (zh), and 40+ subtitle languages
Vimeo (Public Videos)	https://vimeo.com/<id> or https://player.vimeo.com/video/<id>	Extracts existing VTT captions directly (source: "captions"), or falls back to faster-whisper audio transcription when uncaptioned.	English (en), Spanish, German, French, Japanese, Korean, etc.

---

3. Input Example

{
  "videoUrls": [
    "https://www.bilibili.com/video/BV1zUnZ67EMY",
    "https://archive.org/download/call_to_arms_jl_librivox/calltoarms_06_lu.mp3",
    "https://www.ted.com/talks/matt_cutts_try_something_new_for_30_days",
    "https://vimeo.com/1064773380"
  ],
  "languageHint": "auto",
  "useWhisperFallback": true,
  "whisperModel": "small",
  "maxDurationMinutes": 60,
  "maxVideos": 20
}

Input Parameters

Field	Type	Required	Default	Description
videoUrls	array<string>	Yes	—	Publicly accessible URLs from Bilibili (bilibili.com/video/BV...), direct audio links (.mp3/.m4a/.wav), TED Talks (ted.com/talks/...), or public Vimeo videos (vimeo.com/...).
languageHint	string	No	"auto"	Preferred language code ("zh", "en", "ja", etc.) or "auto" for automatic detection.
useWhisperFallback	boolean	No	true	When true, downloads audio and transcribes locally with faster-whisper if no captions exist.
whisperModel	string	No	"small"	Model size for CPU int8 transcription ("base", "small", or "medium").
maxDurationMinutes	integer	No	60	Maximum duration per item in minutes. Videos with metadata duration above this limit are skipped (status: "skipped"), and direct audio files are truncated at this limit.
maxVideos	integer	No	20	Maximum number of URLs processed in one run.

---

4. Output Examples (Real Apify Cloud Outputs)

Example A — Chinese Audio Direct Link (source: "whisper", Archive.org Mandarin Audiobook)

Dataset JSON Record (Excerpt from real cloud run luw3C2TgcZvDlKLOn):

{
  "url": "https://archive.org/download/call_to_arms_jl_librivox/calltoarms_06_lu.mp3",
  "title": "calltoarms_06_lu",
  "durationSec": 262.88,
  "source": "whisper",
  "language": "zh",
  "segments": [
    {
      "start": 0.08,
      "end": 2.64,
      "text": "鲁迅小说集,呐喊"
    },
    {
      "start": 2.9,
      "end": 4.44,
      "text": "一件小事"
    },
    {
      "start": 4.68,
      "end": 7.5,
      "text": "此次 LibriVox 录音由公众所有"
    },
    {
      "start": 9.04,
      "end": 13.4,
      "text": "我从乡下跑到京城里,一转眼已经六年了"
    },
    {
      "start": 13.64,
      "end": 18.52,
      "text": "期间耳闻目睹的所谓国家大事算起来也很不少"
    }
  ],
  "text": "鲁迅小说集,呐喊 一件小事 此次 LibriVox 录音由公众所有 我从乡下跑到京城里,一转眼已经六年了 期间耳闻目睹的所谓国家大事算起来也很不少 ...",
  "srt": "1\n00:00:00,080 --> 00:00:02,640\n鲁迅小说集,呐喊\n\n2\n00:00:02,900 --> 00:00:04,440\n一件小事\n\n3\n00:00:04,680 --> 00:00:07,500\n此次 LibriVox 录音由公众所有\n",
  "srtKey": "SRT_001_calltoarms_06_lu.srt",
  "status": "succeeded",
  "error": null
}

Saved .srt File in Key-Value Store (SRT_001_calltoarms_06_lu.srt):

1
00:00:00,080 --> 00:00:02,640
鲁迅小说集,呐喊

2
00:00:02,900 --> 00:00:04,440
一件小事

3
00:00:04,680 --> 00:00:07,500
此次 LibriVox 录音由公众所有

4
00:00:09,040 --> 00:00:13,400
我从乡下跑到京城里,一转眼已经六年了

5
00:00:13,640 --> 00:00:18,520
期间耳闻目睹的所谓国家大事算起来也很不少

---

Example B — Bilibili Public Mandarin Video (source: "whisper", Real Cloud Run NkJl91DvzPCxh591Y)

{
  "url": "https://www.bilibili.com/video/BV1zUnZ67EMY",
  "title": "陕视评：判决未至，定论先入教材 武大图书馆事件反思：公共出版物不能越司法“断案” （策划：李彤 任超 评论员：刁坤 摄像：姬佩瑶）",
  "durationSec": 176,
  "source": "whisper",
  "language": "zh",
  "segments": [
    {
      "start": 0.02,
      "end": 2.32,
      "text": "一本教大学生心理健康的小采"
    },
    {
      "start": 2.32,
      "end": 4.62,
      "text": "凭什么让一个法院认定清白的男生"
    },
    {
      "start": 4.62,
      "end": 7.42,
      "text": "在白纸黑字上继续当性骚扰犯呢"
    },
    {
      "start": 7.42,
      "end": 9.32,
      "text": "2023年武汉大学图书馆内"
    }
  ],
  "srtKey": "SRT_001_BV1zUnZ67EMY.srt",
  "status": "succeeded",
  "error": null
}

---

Example C — TED Talk / Vimeo Captions Extraction (source: "captions", Real Cloud Run pJBxyhsCk1XEPUI1I)

{
  "url": "https://www.ted.com/talks/matt_cutts_try_something_new_for_30_days",
  "title": "Try something new for 30 days",
  "durationSec": 190,
  "source": "captions",
  "language": "en",
  "segments": [
    {
      "start": 0,
      "end": 4.976,
      "text": "A few years ago, I felt like I was stuck in a rut,"
    },
    {
      "start": 5,
      "end": 6.976,
      "text": "so I decided to follow in the footsteps"
    },
    {
      "start": 7,
      "end": 10.536,
      "text": "of the great American philosopher, Morgan Spurlock,"
    }
  ],
  "srtKey": "SRT_001_1183.srt",
  "status": "succeeded",
  "error": null
}

---

5. Pricing & Cost Examples (Pay-Per-Event)

You only pay for succeeded items based on which path was used:

Event Name	When It Is Charged	Price
captions-transcribed	Charged per video transcribed using existing platform captions or subtitles (source: "captions")	$0.002 / video ($2.00 per 1,000 captioned videos)
whisper-transcribed-minute	Charged per started minute of audio (ceil(durationSec / 60)) transcribed locally via faster-whisper (source: "whisper")	$0.025 / minute

Concrete Cost Examples

• 10 TED Talks or Vimeo videos with existing captions: 10 × $0.002 = $0.020 total (completes in ~10 seconds).
• A 3-minute Bilibili Mandarin video without public captions: 3 minutes × $0.025 = $0.075.
• A 5-minute Chinese audiobook / podcast MP3: 5 minutes × $0.025 = $0.125.
• A 30-minute Chinese podcast episode (MP3): 30 minutes × $0.025 = about $0.75.
• Failed or skipped URLs: $0.00 (never charged).

---

6. Supported vs Unsupported Links

Supported (cloud-verified)

Link type	Example	Expected path	Notes
Bilibili public BV	https://www.bilibili.com/video/BV...	captions (rare) or whisper	Public API audio download; no cookies
Direct audio / podcast	https://.../*.mp3 (Archive.org, RSS CDN, .m4a, .wav)	whisper	Truncated at maxDurationMinutes
TED Talks	https://www.ted.com/talks/...	captions	Official WebVTT / full.vtt
Public Vimeo	https://vimeo.com/<id>	captions or whisper	Rewritten to player.vimeo.com

NOT supported (fail-fast ≤ 2 minutes, no charge)

Link type	Why	Actor behavior
YouTube (with or without captions)	Datacenter IP blocked (Sign in to confirm you're not a bot)	Immediate clear status=failed error — never hangs to TIMED-OUT
Douyin / TikTok share links	Login wall / anti-bot; Actor never uses cookies or paid proxies	Immediate clear status=failed error
Private / VIP / password / DRM	No login, no cookies	failed or skipped with reason

Other limits

• Public Content Only (No Login / No Cookies): Does not support private videos, login-walled pages, VIP/paid Bilibili episodes, password-protected Vimeo videos, or DRM-protected streams.
• Bilibili Unauthenticated Subtitle Restriction: Bilibili requires account login to access AI-generated closed captions on most videos. Since this Actor never logs in or uses cookies, Bilibili videos usually take the whisper audio transcription path.
• CPU Transcription Speed / 10-minute SLA: Whisper runs on Apify Cloud CPU (faster-whisper int8, vad_filter=True, beam_size=1). For audio longer than ~12 minutes the Actor auto-selects base, and above ~25 minutes it selects tiny, so ≤60-minute items can finish within ~10 minutes. Multi-hour files are ffmpeg-cut to maxDurationMinutes before decode to avoid OOM. Caption paths usually finish in seconds.
• Fail-fast timeouts: Metadata / captions / download / whisper each have hard asyncio deadlines; unsupported hosts never call yt-dlp. Per-URL wall clock is capped at 10 minutes.
• Mandarin Homophones in Fast Speech: For rapid news commentary or domain-specific proper nouns without context, whisperModel="small" may occasionally produce same-sound character substitutions (e.g., 统战 → 统政, 小册 → 小采). You can select whisperModel="medium" for higher Mandarin accuracy on complex audio (may miss the 10-minute SLA on long clips).
• Memory & Duration Limits: Requires 2048 MB–4096 MB RAM. Audio exceeding maxDurationMinutes (default 60 minutes) is automatically skipped or truncated at maxDurationMinutes.
• Billing: Actor.charge runs only when status == "succeeded". Failed / skipped / timed-out-internally items cost $0 PPE.

---

7. Frequently Asked Questions (FAQ)

Do I need an OpenAI API key?

No. Transcription runs locally inside the Actor container using faster-whisper (MIT license). No external API key is needed.

Why is YouTube not supported?

YouTube actively blocks requests from cloud datacenter IP addresses unless authenticated cookies or residential proxies are supplied. To keep pricing predictable and compliant, this Actor focuses on Bilibili, podcast/audio direct links, TED Talks, and public Vimeo videos. YouTube URLs fail immediately with a clear error (no TIMED-OUT hang, no PPE charge).

Are Douyin / TikTok links supported?

No. Douyin and TikTok share links require login or anti-bot challenges. They fail within 2 minutes with a clear unsupported-platform message and are never charged.

How can I cap my maximum spend for a run?

Set the Maximum cost per run (ACTOR_MAX_TOTAL_CHARGE_USD) in Apify Console or via the API. The Actor checks your remaining budget before and after each item and stops gracefully as soon as the limit is reached.

Where do I download the .srt subtitle files?

Every succeeded item includes the full .srt text inside the Dataset JSON (srt field) and also saves a standalone .srt file (such as SRT_001_BV1zUnZ67EMY.srt) in the run's Default Key-Value Store.

What open-source licenses are used?

• apify Python SDK: Apache-2.0
• yt-dlp: The Unlicense
• faster-whisper: MIT License
• ffmpeg (system package installed via apt): License (LGPL v2.1+ or GPL v2+/v3) depends on the distribution's compile-time configuration.
• Users are responsible for complying with the Terms of Service of the target platforms they query.